README.md

metadata

dataset_info:
  features:
    - name: original_speech
      dtype: string
    - name: original_language
      dtype: string
    - name: audio_path
      dtype: string
    - name: segment_start
      dtype: float32
    - name: segment_end
      dtype: float32
    - name: transcriptions
      struct:
        - name: de
          dtype: string
        - name: en
          dtype: string
        - name: es
          dtype: string
        - name: fr
          dtype: string
        - name: it
          dtype: string
        - name: nl
          dtype: string
        - name: pl
          dtype: string
        - name: pt
          dtype: string
        - name: ro
          dtype: string
  splits:
    - name: train
      num_bytes: 147857910
      num_examples: 116138
    - name: valid
      num_bytes: 21318985
      num_examples: 17538
    - name: test
      num_bytes: 22580968
      num_examples: 18901
  download_size: 109205144
  dataset_size: 191757863
task_categories:
  - translation
  - text-to-speech
language:
  - es
  - de
  - en
  - fr
  - nl
  - pl
  - pt
  - ro
  - it
size_categories:
  - 100K<n<1M
license: cc-by-nc-4.0

Dataset Card for "Europarl-ST"

Table of Contents

• Table of Contents
• Dataset Description
  • Dataset Summary
  • Languages
• Dataset Structure
  • Data Fields
  • Data Splits
• Additional Information
  • Licensing Information
  • Citation Information

Dataset Description

• Homepage: https://www.mllp.upv.es/europarl-st/
• Paper: https://ieeexplore.ieee.org/document/9054626
• Point of Contact: https://www.mllp.upv.es/

Dataset Summary

Europarl-ST is a Multilingual Speech Translation Corpus, that contains paired audio-text samples for Speech Translation, constructed using the debates carried out in the European Parliament in the period between 2008 and 2012.

Languages

Spanish, German, English, French, Dutch, Polish, Portuguese, Romanian, Italian

Dataset Structure

Data Fields

• original_audio: The original speech that is heard on the recording.
• original_language: The language of the audio
• audio_path: Path to the audio file
• segment_start: Second in which the speech begins
• segment_end: Second in which the speech ends
• transcriptions: Dictionary containing transcriptions into different languages

Data Splits

• train split: 116138 samples
• valid split: 17538 samples
• test split: 18901 samples

Train set (hours):

src/tgt	en	fr	de	it	es	pt	pl	ro	nl
en	-	81	83	80	81	81	79	72	80
fr	32	-	21	20	21	22	20	18	22
de	30	18	-	17	18	18	17	17	18
it	37	21	21	-	21	21	21	19	20
es	22	14	14	14	-	14	13	12	13
pt	15	10	10	10	10	-	9	9	9
pl	28	18	18	17	18	18	-	16	18
ro	24	12	12	12	12	12	12	-	12
nl	7	5	5	4	5	4	4	4	-

Valid/Test sets are all between 3 and 6 hours.

Additional Information

Licensing Information

• The work carried out for constructing the Europarl-ST corpus is released under a Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0)

• All rights of the data belong to the European Union and respective copyright holders.

Citation Information

If you use the corpus in your research please cite the following reference:

@INPROCEEDINGS{jairsan2020a, author={J. {Iranzo-Sánchez} and J. A. {Silvestre-Cerdà} and J. {Jorge} and N. {Roselló} and A. {Giménez} and A. {Sanchis} and J. {Civera} and A. {Juan}}, booktitle={ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, title={Europarl-ST: A Multilingual Corpus for Speech Translation of Parliamentary Debates}, year={2020}, pages={8229-8233},}